Does anyone notice the AI has a higher rate of interruption when a user starts giving numbers in their response? For example, the caller gives an address or a callback number? I originally had my LLM request delay at 0.1 seconds and increased to 0.5, but it's still happening quite frequently. Doing roughly 2000-4000 inbound minutes daily and this has been a noticeable comlpaint that we're trying to analyze and address properly. Any suggestions? - Using Deepgram Nova 2 and Eleven Labs